40occupations with a baseline AI-exposure score lifted from Goldman Sachs, the WEF, McKinsey and the OECD. Then a live feed: every article we process gets classified against this taxonomy — which roles the news is framing as displaced, augmented, or quietly creating new work.
Each dot is one occupation, plotted left-to-right by its baseline AI-exposure score. Bright dotsare roles the news is talking about this week — colour shows the dominant framing. Faint outlines are quiet this week.
The three roles getting the heaviest displacement framing in the last seven days. Sample evidence pulled straight from the article that earned each tag.
“Anthropic cut the share of junior technical hires as AI takes over more work; one-fifth of software engineering managers reported reducing hiring because of AI.”Anthropic Cuts Share Of Junior Technical Hires As AI Takes Over More Work - BW P… →
Active roles — with coverage this week — show their sentiment split and the article that earned each tag. Quiet roles are listed at the foot of each section so the taxonomy stays complete without burying the signal.
No coverage in this category this week.
No coverage in this category this week.
No coverage in this category this week.
Application development, systems work, feature delivery.
No coverage in this category this week.
No coverage in this category this week.
No coverage in this category this week.
No coverage in this category this week.
The taxonomy is fixed: 40occupations across eight categories, chosen to span the full range of arguments in the literature — from “near-total automation” (telemarketers, data-entry clerks) to “barely touched for a decade” (plumbers, surgeons). Each role carries a baseline risk score lifted from published research, with the source cited on the role’s row so you can check our working.
These baselines are task-exposurefigures — the share of a job’s tasks that today’s models could plausibly automate. That’s not the same as “% of jobs that will disappear”. Goldman’s original paper is explicit about this: high exposure can mean augmentation, lower labour intensity, or displacement, depending on how firms respond. We treat the score as a prior, not a prediction.
Every article the AI Risk Clock processes gets sent to DeepSeek alongside the full taxonomy. The model returns a short JSON list of which roles the article meaningfully discussesin connection with AI — not passing mentions, not inferred categories — plus a sentiment tag per role:
Each mention carries a short evidence snippet quoting or paraphrasing the line that earned the tag. The per-role counts shown above are rolling 14-day totals; the sentiment bar is the last 7 days only.